Classic video games are notoriously difficult, coming from an era where the primary inspirations were arcade games, which were designed to earn as many quarters as possible from those that played them. Things have gotten quite a bit easier since then in most cases, but the classics are still there to challenge gamers whenever they feel the need. Or, in some cases, as a test for just how great AI can be.
Scientists at the Uber AI Labs have been teaching AI to beat old school video games, including the like’s of iconic Atari platformer Pitfall. The methodology behind this is fairly interesting, with researchers creating algorithms that allow the AI to remember areas that it’s previously explored, helping it keep track of where it’s been, and using that information to push it to new locations.
Older methods, specifically one called “Intrinsic Motivation” rewarded AI for finding new areas, but could result in AI forgetting areas it had previously discovered, and by extension, any areas that it might have discovered through previously-forgotten routes. That issue is known as “detachment,” but with the new method, AI has not only been able to complete Pitfall, but also the Atari games Freeway, and Montezuma’s Revenge. Essentially, the new algorithms have weeded out the “random action” element.
An AI beating classic arcade games may sound pointless, but it has practical applications. The goal is to create AI that’s better at navigating real-world environments, something that digital environments can help it train for. This is particularly helpful for things like real-world disaster zones, where AI would be able to help robots search for survivors after an earthquake, hurricane, or any other calamity. It’s yet another example of video games having a benefit in day-to-day life, though the melding of AI and video games seems like something out of a sci-fi film.
Obviously, this method has a lot of potential, both for grand scheme problems and smaller, in-home applications. One example provided was asking a robot to get the user a cup of coffee. A robot wouldn’t be able to do that through random actions like older AI used, rather it would be a learning process.
AI and video games operate at an interesting intersection, and this isn’t the first time AI has been used in a digital space. While AI within video games is frequently criticized for acting unintelligent, seeing games utilized to make smarter flips the issue on its head. Hopefully, this results in smarter AI all around, but time will tell.
Source: BBC
![\"IT電腦補習](\"http://itteacheritfreelance.hk/test/wordpress/wp-content/uploads/2016/05/logo2-2.png\"){kind=logo}
www.itsec.hk
www.itsec.vip
www.itseceu.uk
Leave a Reply